# Bring Split's wall layer onto the piece model · dara/icarus

[View on GitCafe](https://git.cafe/dara/icarus/commit/87a2144b2156630a5ceb578afd8585ea5340f5bb)

Repository: [dara/icarus](https://git.cafe/dara/icarus)

Visibility: public

Requested revision: 87a2144b2156630a5ceb578afd8585ea5340f5bb

Requested commit: 87a2144b2156630a5ceb578afd8585ea5340f5bb

Commit: 87a2144b2156630a5ceb578afd8585ea5340f5bb

Tree: ed4a810cb7b9fbe7099304324873a596e76401a5

Author: Dara Adedeji

Committer: Dara Adedeji

## Message

```
Bring Split's wall layer onto the piece model

Split was the prototype: 69 wall records, one per painted run, each with a
single hand-assigned band, while the other twelve maps carry thousands of
pieces with bands measured from the 3D scene. Every rule on this branch is
piece-based, so Split was skipped by all of them.

scripts/partition_split_walls.py cuts each record into pieces about two SVG
units long along its medial line, keeping the parent's floor and bands so
the cut alone changes nothing; named prop cuts stay whole and protected.
The pieces then take the same path as the other maps: ray derivation,
reviewed-wall protection, the narrow-footprint raise cap, the perimeter
seal (its Split exemption removed), anomaly fixes and notch closing.
Neighbour smoothing is skipped for Split because its baseline bands are
upper bounds, not reviews. 3540 attack and 3520 defense pieces result.

The notch rule now also closes a piece the probe left with no band at all
between two floor-anchored neighbours, which touched a handful of pieces on
the other maps. The legacy Split sightline test accepts piece ids under the
parent walls it names.

Gate: all tests pass.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

```

## Parents

- [d6337deebf4404c9a7ea0b8b2727fce6377aeda0](https://git.cafe/dara/icarus/commit/d6337deebf4404c9a7ea0b8b2727fce6377aeda0?format=markdown)

[Source at this commit](https://git.cafe/dara/icarus/tree/87a2144b2156630a5ceb578afd8585ea5340f5bb?format=markdown)